Michael Hoffmann, Jophin John, Stefan Schweter
et al.
We present Llama-GENBA-10B, a trilingual foundation model addressing English-centric bias in large language models. Built on Llama 3.1-8B and scaled to 10B parameters, Llama-GENBA-10B is continuously pretrained on 164B tokens (82B English, 82B German, and 80M Bavarian), balancing resources while preventing English dominance. Targeted at the German NLP community, the model also promotes Bavarian as a low-resource language. Development tackled four challenges: (1) curating a multilingual corpus despite Bavarian scarcity, (2) creating a unified tokenizer for English, German, and Bavarian, (3) optimizing architecture and language-ratio hyperparameters for cross-lingual transfer, and (4) establishing the first standardized trilingual evaluation suite by translating German benchmarks into Bavarian. Evaluations show that Llama-GENBA-10B achieves strong cross-lingual performance, with the fine-tuned variant surpassing Apertus-8B-2509 and gemma-2-9b in Bavarian and establishing itself as the best model in its class for this language, while also outperforming EuroLLM in English and matching its results in German. Training on the Cerebras CS-2 demonstrated efficient large-scale multilingual pretraining with documented energy use, offering a blueprint for inclusive foundation models that integrate low-resource languages.
Tabia Tanzin Prama, Chhandak Bagchi, Vishal Kalakonnavar
et al.
This study examines whether German X users would see politically balanced news feeds if they followed comparable leading politicians from each federal parliamentary party of Germany. We address this question using an algorithmic audit tool [1] and all publicly available posts published by 436 German politicians on X. We find that the default feed of X showed more content from far-right AfD than from other political parties. We analyze potential factors influencing feed content and the resulting political non-representativeness of X. Our findings suggest that engagement measures and unknown factors related to party affiliation contribute to the overrepresentation of extremes of the German political party spectrum in the default algorithmic feed of X.
Within this study, the influence of the German Information Retrieval community on the retrieval campaigns Text Retrieval Conference (TREC) and Conference and Labs of the Evaluation Forum (CLEF) between 2000 and 2022 was analyzed based on metadata provided by OpenAlex and further metadata extracted with the GROBID framework from the publication's full texts. The analysis was conducted at the institutional and researcher levels. It was found that the German IR community, both on the author and institution level, mainly contributed to CLEF. Furthermore, it was shown that productivity follows the assumptions made by Lotka's Law.
We combine the theory of inductive data types with the theory of universal measurings. By doing so, we find that many categories of algebras of endofunctors are actually enriched in the corresponding category of coalgebras of the same endofunctor. The enrichment captures all possible partial algebra homomorphisms, defined by measuring coalgebras. Thus this enriched category carries more information than the usual category of algebras which captures only total algebra homomorphisms. We specify new algebras besides the initial one using a generalization of the notion of initial algebra.
Typically 20–40 extreme cyclone events (sometimes called ‘weather bombs’) occur in the Arctic North Atlantic per winter season, with an increasing trend of 6 events/decade over 1979–2015, according to 6 hourly station data from Ny-Ålesund. This increased frequency of extreme cyclones is consistent with observed significant winter warming, indicating that the meridional heat and moisture transport they bring is a factor in rising temperatures in the region. The winter trend in extreme cyclones is dominated by a positive monthly trend of about 3–4 events/decade in November–December, due mainly to an increasing persistence of extreme cyclone events. A negative trend in January opposes this, while there is no significant trend in February. We relate the regional patterns of the trend in extreme cyclones to anomalously low sea-ice conditions in recent years, together with associated large-scale atmospheric circulation changes such as ‘blockinglike’ circulation patterns (e.g. Scandinavian blocking in December and Ural blocking during January–February).
AbstractThe paper presents examples of meta-morphomes (a kind of morphomic patterns, involving syncretisms) in North Germanic. There has been some debate over the notion of such patterns, and the aim is therefore to present relatively clear cases. Five cases are presented, involving inflection in verbs, nouns and adjectives. The syncretisms are all ‘unnatural’; they do not make much sense for syntax, semantics or phonology. While patterns that are obvious to the linguist are not necessarily obvious to speakers, the paper presents diachronic evidence that these morphomic patterns have been noticed by speakers. At least some criticism against ‘morphomic’ analyses is based on implausible premises: An analysis in terms of features is not automatically preferable only by being possible; the idea of ‘taking morphology seriously’ is untenable; the claim that the morphomic approach is a mere enumeration of facts may involve a self-contradiction.
Neslihan Iskender, Oleg Vasilyev, Tim Polzehl
et al.
Evaluating large summarization corpora using humans has proven to be expensive from both the organizational and the financial perspective. Therefore, many automatic evaluation metrics have been developed to measure the summarization quality in a fast and reproducible way. However, most of the metrics still rely on humans and need gold standard summaries generated by linguistic experts. Since BLANC does not require golden summaries and supposedly can use any underlying language model, we consider its application to the evaluation of summarization in German. This work demonstrates how to adjust the BLANC metric to a language other than English. We compare BLANC scores with the crowd and expert ratings, as well as with commonly used automatic metrics on a German summarization data set. Our results show that BLANC in German is especially good in evaluating informativeness.
Karolina Zaczynska, Nils Feldhus, Robert Schwarzenberg
et al.
Pre-trained transformer language models (TLMs) have recently refashioned natural language processing (NLP): Most state-of-the-art NLP models now operate on top of TLMs to benefit from contextualization and knowledge induction. To explain their success, the scientific community conducted numerous analyses. Besides other methods, syntactic agreement tests were utilized to analyse TLMs. Most of the studies were conducted for the English language, however. In this work, we analyse German TLMs. To this end, we design numerous agreement tasks, some of which consider peculiarities of the German language. Our experimental results show that state-of-the-art German TLMs generally perform well on agreement tasks, but we also identify and discuss syntactic structures that push them to their limits.
Over the last two decades, the majority of German universities adopted various characteristics of the prevailing North-American academic system, resulting in significant changes in several key areas that include, e.g., both teaching and research. The universities' internal organizational structures, however, still follow a traditional, decentralized scheme implementing an additional organizational level -- the Chair -- effectively a "mini department" with dedicated staff, budget and infrastructure. Although the Technical University of Munich (TUM) has been establishing a more centralized scheme for many administrative tasks over the past decade, the transition from its distributed to a centralized information technology (IT) administration and infrastructure is still an ongoing process. In case of the authors' chair, this migration so far included handing over all network-related operations to the joint compute center, consolidating the Chair's legacy server system in terms of both hardware architectures and operating systems and, lately, moving selected services to replacements operated by Department or University. With requirements, individuals and organizations constantly shifting, this process, however, is neither close to completion nor particularly unique to TUM. In this paper, we will thus share our experiences w.r.t. this IT migration as we believe both that many of the other German universities might be facing similar challenges and that, in the future, North-American universities - currently not implementing the chair layer and instead relying on a centralized IT infrastructure - could need a more decentralized solution. Hoping that both benefit from this journey, we thus present the design, commissioning and evolution of our infrastructure.
Paper I described the normal features of the North Equatorial Belt (NEB) in recent years, especially the large dark formations which are thought to represent waves in the prograde jet on the NEB south edge (NEBs, 7N), and the NEB expansion events (NEEs) in which the belt broadens to the north at intervals of 3 to 5 years. Here I describe an exceptional set of apparently coordinated changes which occurred in 2011-12, after more localised precursors in 2008 and 2010. (1) The large NEBs dark formations progressively disappeared until none remained. (2) In the sectors of NEBs thus vacated, smaller dark features all moved with unprecedented super-fast speeds, which were modulated by the few normal features as long as they lasted, and then accelerated further, reaching 139-151 m/s. (3) Rifts (expanding systems of convective white clouds) also disappeared from the NEB. (4) The NEB north half progressively faded (whitened) until there was only a narrow, southerly NEB, narrower and fainter than it had been for nearly a century. These changes have several profound implications for understanding the dynamics of the region. First, the NEBs took on the same appearance, dynamics, and speed, as the equivalent jet at 7S (SEBn), supporting the view that the two jets are essentially symmetrical, with an underlying jet in the range 150-170 m/s. Secondly, the manifestation of this jet at the surface is normally suppressed by the presence of large slow-moving formations, which are probably Rossby waves. Thirdly, the loss of the large dark formations and the narrowing of the belt may have been promoted by the decline of convective rift activity. Fourthly, these changes seem to represent a reversion to the situation that existed before 1912, when such appearances were common and were followed cyclically by vigorous NEB Revivals. Indeed, just such a Revival would ensue in 2012.
This article presents a synopsis of the activity in Jupiter's North Equatorial Belt (NEB) from 1986 to 2010, and of the speeds of dark formations on its south edge and bright streaks ('rifts') in its interior. In particular I discuss NEB expansion events (NEEs), which took place every 3-5 years during this time, and how the various features of the NEB are involved in them. I present evidence that the NEE affects not just the northern edge, but the whole width of the belt. It begins with an outbreak of a bright rift that is more northerly and slower-moving than usual; this is often involved with the first ejection of dark material northwards into the N. Tropical Zone, but typically the rift also expands southwards across the width of the NEB. NEBs dark formations are usually affected, as they are during individual interactions with rifts at other times; they may be disrupted, or intensified, and they usually undergo deceleration. The expansion of the dark NEB to the north occurs concurrently, and is followed by the appearance of new dark 'barges' and white ovals flanking the NEBn jet. The speed of the NEBs dark formations varies with their mean spacing, consistent with the prevailing hypothesis that they are planetary Rossby waves. In most apparitions since 2000 we have also detected smaller, faster features (120 m/s). I propose that these represent waves of the same type, but with higher frequency, and that their speed is slightly less than the true wind speed at cloud-top level under normal conditions.
We present a photometric catalog for Spitzer Space Telescope warm mission observations of the North Ecliptic Pole (NEP; centered at $\rm R.A.=18^h00^m00^s$, $\rm Decl.=66^d33^m38^s.552$). The observations are conducted with IRAC in 3.6 $μ$m and 4.5 $μ$m bands over an area of 7.04 deg$^2$ reaching 1$σ$ depths of 1.29 $μ$Jy and 0.79 $μ$Jy in the 3.6 $μ$m and 4.5 $μ$m bands respectively. The photometric catalog contains 380,858 sources with 3.6 $μ$m and 4.5 $μ$m band photometry over the full-depth NEP mosaic. Point source completeness simulations show that the catalog is 80% complete down to 19.7 AB. The accompanying catalog can be utilized in constraining the physical properties of extra-galactic objects, studying the AGN population, measuring the infrared colors of stellar objects, and studying the extra-galactic infrared background light.
Due to the rapid increase of digitization within our society, digital identities gain more and more importance. Provided by the German eID solution, every citizen has the ability to identify himself against various governmental and private organizations with the help of his personal electronic ID card and a corresponding card reader. While there are several solutions available for desktop use of the eID infrastructure, mobile approaches have to be payed more attention. In this paper we present a new approach for using the German eID concept on an Android device without the need of the actual identity card and card reader. A security evaluation of our approach reveals that two non-critical vulnerabilities on the architecture can't be avoided. Nevertheless, no sensitive information are compromised. A proof of concept shows that an actual implementation faces some technical issues which have to be solved in the future.
The present article is a summary of the book English: The Language of the Vikings by Joseph E. Emonds and Jan Terje Faarlund. The major claim of the book and of this article is that there are lexical and, above all, syntactic arguments in favor of considering Middle and Modern English as descending from the North Germanic language spoken by the Scandinavian population in the East and North of England prior to the Norman Conquest, rather than from the West Germanic Old English.
Pleistocene climate changes have imposed extreme conditions to intertidal rocky marine communities, forcing many species to significant range shifts in their geographical distributions. Phylogeographic analyses based on both mitochondrial and nuclear genetic markers provide a useful approach to unravel phylogeographic patterns and processes of species after this time period, to gain general knowledge of how climatic changes affect shifts in species distributions. We analyzed these patterns on the corkwing wrasse (Symphodus melops, Labridae), a rocky shore species inhabiting North Sea waters and temperate northeastern Atlantic Ocean from Norway to Morocco including the Azores, using a fragment of the mitochondrial control region and the first intron of the nuclear S7 ribosomal protein gene. We found that S. melops shows a clear differentiation between the Atlantic and the Scandinavian populations and a sharp contrast in the genetic diversity, high in the south and low in the north. Within each of these main geographic areas there is little or no genetic differentiation. The species may have persisted throughout the last glacial maximum in the southern areas as paleotemperatures were not lower than they are today in North Scandinavia. The North Sea recolonization most likely took place during the current interglacial and is dominated by a haplotype absent from the south of the study area, but present in Plymouth and Belfast. The possibility of a glacial refugium in or near the English Channel is discussed.
We study the number density distribution of a sample of K and M dwarf stars, matched North and South of the Galactic plane within a distance of 2 kpc from the sun, using observations from the Ninth Data Release of the Sloan Digital Sky Survey. We determine distances using the photometric parallax method, and in this context systematic effects exist which could potentially impact the determination of the number density profile with height from the Galactic plane --- and ultimately affect a number density North-South asymmetry. They include: (i) the calibration of the various photometric parallax relations, (ii) the ability to separate dwarfs from giants in our sample, (iii) the role of stellar population differences such as age and metallicity, (iv) the ability to determine the offset of the sun from the Galactic plane, and (v) the correction for reddening from dust in the Galactic plane, though our stars are at high Galactic latitudes. We find the various analyzed systematic effects to have a negligible impact on our observed asymmetry, and using a new and larger sample of stars we confirm and refine the earlier discovery of Widrow et al. of a significant Galactic North-South asymmetry in the stellar number density distribution.